Mcp servers
MCP Servers
Tool servers that expose your infrastructure to AI
MCP (Model Context Protocol) is an open standard for connecting AI models to external tools, data sources, and services. GNETiX uses MCP as the bridge between the AI Director and your infrastructure.
An MCP server exposes three types of capabilities:
| Capability | Description |
|---|---|
| Tools | Functions the AI can call (e.g., show_interface_status, restart_pod, query_metrics) |
| Resources | Data the AI can read (e.g., configuration files, documentation, runbooks) |
| Prompts | Reusable prompt templates for common operations |
How MCP Fits Into GNETiX
User Message --> Director (LLM) --> Tool Call --> On-Prem Agent --> MCP Server --> Infrastructure
^
Streamable HTTP- The Director decides which tool to call based on the user's request
- The tool call is dispatched to an on-prem sub-agent via the WebSocket relay
- The sub-agent connects to the appropriate MCP server over Streamable HTTP
- The MCP server executes the tool against your infrastructure and returns the result
- The result flows back to the Director for synthesis
MCP servers run on-prem alongside your infrastructure. They never need to be exposed to the internet. The sub-agent handles all communication with the cloud via outbound WebSocket.
Included Example Servers
GNETiX ships with reference MCP server implementations that you can use as starting points:
| Server | Description | Technology |
|---|---|---|
| network-mcp | Network device management (show commands, config changes) | Netmiko |
| kubernetes-mcp | Kubernetes cluster operations (pods, deployments, logs) | kubernetes-client |
| dynatrace-mcp | Monitoring data retrieval (problems, metrics, entities) | Dynatrace API |
| cribl-mcp | Log pipeline management | Cribl API |
Each example server is a standalone Python project with a Dockerfile, pyproject.toml, and well-documented tool definitions.
Next Steps
- Building MCP Servers -- create your own custom tool server
- Tool Catalog -- manage, tag, and secure tools across all MCP servers